一些股票代码的留存

#提前缩小数据量
df_t = df1.loc[(df1['rownum_W'] <= (sst+25))]
t = df_t.groupby(['rownum_W'],as_index=False).ts_code.count()
row_liu = t['rownum_W'].loc[t.ts_code>12000]

df1 = pd.DataFrame()
for row in row_liu:
    df_for = df_t[df_t.rownum_W==row]
    df1 = pd.concat([df1,df_for],axis=0)

#防止集中某一天涨的多，抽取后再打乱数据
df_1 = pd.DataFrame()
df_0 = pd.DataFrame()

for dt in df1.date.unique():
    qx_1 = df1_1[df1_1.date==dt]
    if len(qx_1)<1250:
        qx1 = qx_1[qx_1.date == dt]
    else:
        qx1 = qx_1[qx_1.date == dt].sample(n=1250)
    df_1 = pd.concat([df_1,qx1],axis=0)

for dt in df1.date.unique():
    qx_0 = df1_0[df1_0.date==dt]
    if len(qx_0)<3000:
        qx0 = qx_0[qx_0.date == dt]
    else:
        qx0 = qx_0[qx_0.date == dt].sample(n=3000)
    df_0 = pd.concat([df_0,qx0],axis=0)

df1 = pd.concat([df_1,df_0],axis=0)

from sklearn.utils import shuffle
df1 = shuffle(df1)